Machine Learning for Cardiovascular Disease Prediction: Global Research Trends and Comparative Assessment

Abstract

Cardiovascular disease (CVD) prediction has become an active area of machine learning research, yet publication trends and empirical model behavior are often examined separately. This study integrates bibliometric mapping with empirical machine learning evaluation to examine recent developments in CVD prediction. A corpus of 3,171 Scopus-indexed publications published between 2021 and 2026 was analyzed to identify publication growth, citation patterns, collaboration networks, influential documents, keyword clusters, and thematic evolution. The bibliometric findings provided context for the subsequent empirical comparison of representative models using structured clinical data and explainability analysis. The experimental component used $\mathbf{6 8, 0 9 8}$ patient records after preprocessing and compared Extreme Gradient Boosting, Random Forest, Logistic Regression, and Linear Support Vector Machine. Extreme Gradient Boosting achieved the highest ROC-AUC among the evaluated models, although its performance was only marginally higher than Random Forest. SHapley Additive exPlanations highlighted systolic blood pressure as the most influential predictor in the final model, followed by age and cholesterol level. The findings suggest increasing attention toward clinically oriented and interpretable prediction models, while external validation remains necessary before practical deployment.